Project Summary Safety net providers treat a substantial share of socioeconomically vulnerable patients in their communities, but struggle to provide timely access to high quality specialty care for their patients. Delayed access to specialty care is associated with worse health outcomes and potentially contributes to health disparities across socioeconomic groups. Given their limited resources, safety net providers must seek creative approaches to improve specialty access. However, to choose what programs to implement, safety net providers need to understand the specialty care needs of their populations. Fortunately, the adoption of eConsult systems by safety net providers across the US provides a valuable opportunity to systematically measure patterns of specialty care referrals for minority, underserved populations. In this project, we propose using state-of-the-art methods in machine learning and natural language processing (NLP) to help safety net providers extract actionable, population wide data from their electronic consultation systems. We will do this in partnership with three of the most prominent safety net health systems in the US in Los Angeles, San Francisco and New York City. Using specialty request databases from our collaborators, we will build NLP systems to automatically classify specialty requests along two dimensions: the ?clinical issue? motivating the request (e.g., chest pain), and the ?question type? (e.g., request for a procedure, help with medication management). This automated classification of electronic specialty requests can enable identification of promising targets for interventions to improve specialty access and quality of care. After developing these NLP systems, we will analyze >1 million specialty requests to describe trends in how safety net patients are referred to specialists and examine variation in referral patterns by clinic and individual provider. The goal is to identify the most impactful opportunities to improve specialty access and quality. For example, a high rate of referrals for esophageal reflux, which most PCPs can treat on their own with specialist guidance, could lead to new treatment algorithms, potentially reducing the need for these requests and improving access for other patients. This proposal is a ?high-risk high-reward? project that creates new research tools to identify and evaluate data-driven interventions to improve specialty care delivery for underserved populations.